FeaturedAssume the Model is Compromised
Input defenses will never catch every hostile prompt, so this paper argues for the alternative: build systems where a compromised model can't do damage. A practical guide to sandboxing, isolation, and bounding the blast radius of agent runs, with evidence to prove it.
Read article
There Is No Firewall for a Sentence
You cannot secure a language model the way you secure software: it reads instructions and untrusted data as one stream, and nothing reliably tells them apart. Here is what building LuumenAI taught us about the one layer where you still have control.
Drop the Backpack: What $900/Day in AI Costs Taught Us About MCP
MCP bleeds tokens: tool definitions load upfront and every call compounds context. How we cut LuumenAI's costs 98% with sub-agents and code execution.